Using graph neural networks to forecast the top 40 index in finance
| dc.contributor.author | Moepi, Johannah Reabetswe | |
| dc.date.accessioned | 2022-09-13T06:47:05Z | |
| dc.date.available | 2022-09-13T06:47:05Z | |
| dc.date.issued | 2021 | |
| dc.description | A research report submitted in partial fulfillment of the requirements for the degree of Master of Science in the field of e-Science in the School of Computer Science and Applied Mathematics, University of the Witwatersrand, 2021 | en_ZA |
| dc.description.abstract | Context: After various researches and gauging the amount of work put in fore-casting techniques, not much work is done in graph classification task where not only the stock market is predicted but individual stock prices and market index movements can be predicted together. In this thesis, the work will be based on capturing the effects of the South African Volatility Index (SAVI) and FTSE/JSE Top 40 index on the stock market volatility. Aims: Is to develop a graph neural network model that forecasts the Top 40 index, to use a different technique and to evaluate what happens when the two fore-casts are put together. Method: The data is simulated in two steps. For SAVI, there is data period of 2007 February 3 to 2012 July 28. The relation data consisting of FTSE/JSE Top 40 constituents is from 2019 December 1 to 2020 March 31. The data was normalized through window sizing and LSTM model was implemented for stock market pre-diction. T-distributed stochastic neighbor embedding (t-SNE) algorithm is used for nodes representation. For prediction of SAVI, the volatility data is being stationarized first then LSTM model is applied to forecast it. Results: The data set was divided into training and testing data. Using LSTM method and historical volatility models , the stock market was predicted with a dramatic fall when there was high volatility. A t-SNE map visualized the node representations and then classified the companies into their specific branches . The effect of using different relation data was evaluated and stock market was predicted to see if the data has any impact on it. SAVI data was stationarized and forecasted by LSTM technique. Conclusion: Always normalize and balance the data to avoid biased incomparable results. The implemented LSTM model predicted the stock price and volatility. For implementation of a hierarchical attention network, node representations of the relation data were required and implemented. With different relations in the data ,the effect of using multiple relations in the stock market prediction was studied and poor prediction of the stock market was obtained. As such a hierarchical attention network was not able to be taken any further | en_ZA |
| dc.description.librarian | CK2022 | en_ZA |
| dc.faculty | Faculty of Science | en_ZA |
| dc.identifier.uri | https://hdl.handle.net/10539/33158 | |
| dc.language.iso | en | en_ZA |
| dc.school | School of Computer Science and Applied Mathematics | en_ZA |
| dc.title | Using graph neural networks to forecast the top 40 index in finance | en_ZA |
| dc.type | Thesis | en_ZA |